Hybrid approach for query recommendation in conversation systems

• US 20070299824A1
• Filed: 06/27/2006
• Published: 12/27/2007
• Est. Priority Date: 06/27/2006
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Pan, Shimei, Shaw, James C.
• Granted Patent: US 7,533,089 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06F 40/56 Natural language generation
• CPC Class Codes: Y10S 707/99934 Query formulation, input preparation, or translation
• CPC Class Codes: Y10S 707/99935 Query augmenting and refining, e.g. inexact access

Abstract

Techniques are disclosed for combining natural language generation with query retrieval for context appropriate query recommendation. For example, a computer-implemented method for generating a recommended query for a conversation system in response to an original user query, wherein at least a portion of the original user query is not understandable to a query interpretation process, includes the following steps. Recommendation results are computed, in response to the original user query, using a natural language generation-based recommendation process. Recommendation results are computed, in response to the original user query, using a retrieval-based recommendation process. A recommended query is generated based on consideration of at least a portion of the natural language generation-based recommendation results and at least a portion of the retrieval-based recommendation results.

Claims

1. A computer-implemented method for generating a recommended query for a conversation system in response to an original user query, wherein at least a portion of the original user query is not understandable to a query interpretation process, the method comprising the steps of:

computing recommendation results, in response to the original user query, using a natural language generation-based recommendation process;

computing recommendation results, in response to the original user query, using a retrieval-based recommendation process; and

generating a recommended query based on consideration of at least a portion of the natural language generation-based recommendation results and at least a portion of the retrieval-based recommendation results.

2. The method of claim 1, wherein the natural language generation-based recommendation result computing step comprises the step of selecting content for use in generating the recommended query.

3. The method of claim 2, wherein the content selection step comprises the step of extracting one or more features from at least one of the original user query and interpretation results generated by the query interpretation process, wherein the one or more extracted features characterize a current interpretation problem associated with the original user query.

4. The method of claim 3, wherein the content selection step further comprises, based on the one or more extracted features, determining and applying one or more rules for use in revising at least a portion of the original user query to generate semantic content for use in generating the recommended query.

5. The method of claim 4, wherein the natural language generation-based recommendation result computing step further comprises, based on the semantic content, generating a grammatically-appropriate sentence which forms the recommended query.

6. The method of claim 5, wherein the grammatically-appropriate sentence generating step further comprises retrieving one or more of words, phrases and sentence segments from at least one of the original user query and a query corpus to convey the semantic content for the recommended query.

7. The method of claim 1, wherein the retrieval-based recommendation result computing step comprises the step of computing one or more similarity scores based on the original user query and one or more example queries from a query corpus.

8. The method of claim 7, wherein the one or more similarity scores comprise a surface similarity score.

9. The method of claim 7, wherein the one or more similarity scores comprise a semantic similarity score.

10. The method of claim 7, wherein the one or more similarity scores comprise a context similarity score.

11. The method of claim 7, wherein the one or more similarity scores are used to identify one or more of the example queries from the query corpus to be included as part of the retrieval-based recommendation results.

12. Apparatus for generating a recommended query for a conversation system in response to an original user query, wherein at least a portion of the original user query is not understandable to a query interpretation process, the apparatus comprising:

a memory; and

at least one processor coupled to the memory and operative to;

(i) compute recommendation results, in response to the original user query, using a natural language generation-based recommendation process;

(ii) compute recommendation results, in response to the original user query, using a retrieval-based recommendation process; and

(iii) generate a recommended query based on consideration of at least a portion of the natural language generation-based recommendation results and at least a portion of the retrieval-based recommendation results.

13. The apparatus of claim 12, wherein the natural language generation-based recommendation result computing operation comprises the step of selecting content for use in generating the recommended query.

14. The apparatus of claim 13, wherein the content selection operation comprises extracting one or more features from at least one of the original user query and interpretation results generated by the query interpretation process, wherein the one or more extracted features characterize a current interpretation problem associated with the original user query.

15. The apparatus of claim 14, wherein the content selection operation further comprises, based on the one or more extracted features, determining and applying one or more rules for use in revising at least a portion of the original user query to generate semantic content for use in generating the recommended query.

16. The apparatus of claim 15, wherein the natural language generation-based recommendation result computing operation further comprises, based on the semantic content, generating a grammatically-appropriate sentence which forms the recommended query.

17. The apparatus of claim 16, wherein the grammatically-appropriate sentence generating operation further comprises retrieving one or more of words, phrases and sentence segments from at least one of the original user query and a query corpus to convey the semantic content for the recommended query.

18. The apparatus of claim 12, wherein the retrieval-based recommendation result computing operation comprises computing one or more similarity scores based on the original user query and one or more example queries from a query corpus.

19. The apparatus of claim 18, wherein the one or more similarity scores comprise one or more of a surface similarity score, a semantic similarity score, and a context similarity score.

20. An article of manufacture for generating a recommended query for a conversation system in response to an original user query, wherein at least a portion of the original user query is not understandable to a query interpretation process, comprising a machine readable medium containing one or more programs which when executed implement the steps of:

computing recommendation results, in response to the original user query, using a natural language generation-based recommendation process;

computing recommendation results, in response to the original user query, using a retrieval-based recommendation process; and

generating a recommended query based on consideration of at least a portion of the natural language generation-based recommendation results and at least a portion of the retrieval-based recommendation results.